Multiple outlier detection in multivariate data using self-organizing maps (Q2488397)

The Self-Organizing Map (SOM) is a two-layer feed forward neural network that uses an unsupervised training algorithm for the (nonlinear) projection of multivariate data onto a two-dimensional lattice (rectangular or hexagonal) grid. The authors propose to use the resulting projections for outlier detection. The candidates for outliers are the data, which were projected on a grid cell lying far from its neighbours. The performance of this technique is assessed on simulated and real life data.